When developing a network infrastructure, one usually considers either local computing or cloud computing. But these two options and their combinations are not enough. For example, what to do if you cannot refuse cloud computing, but there is not enough bandwidth or the traffic is too expensive?
Add an intermediate link that will perform part of the calculations at the edge of the local network or production process. This peripheral concept is called Edge Computing. The concept complements the current cloud data usage model, and in this article we'll look at the required hardware and sample tasks for it.
edge computing levels
Let's say you have a whole bunch of sensors installed in your house: a thermometer, a hygrometer, a light sensor, leaks, and so on. The logic controller processes the information coming from them, implements automation scripts, issues the processed telemetry to the cloud service and receives updated automation scripts and fresh firmware from it. Thus, local calculations are performed directly on the object, but the equipment is controlled from a node that combines many such devices.
This is an example of a very simple edge computing system, but it already shows all three levels of edge computing:
- IoT devices: generate "raw data" and transmit it using various protocols.
- Edge nodes: process data in close proximity to information sources and act as temporary data stores.
- Cloud services: offer management functions for both edge and IoT devices, perform long-term data storage and analysis. In addition, they support integration with other corporate systems.
The concept of edge computing itself is part of a large ecosystem that optimizes the workflow. It includes both hardware (rack and edge servers) and network and software parts (for example, the platform for the development of AI algorithms). Since bottlenecks can form during the creation, transfer and processing of big data and limit the performance of the entire system, these parts must be compatible with each other.
Edge Server Features
At the level of peripheral nodes, Edge Computing uses edge servers that are placed directly where information is produced. Usually these are production or technical premises in which it is impossible to install a server rack and ensure cleanliness. Thus, edge servers are implemented in compact dust- and moisture-proof cases with an extended temperature range; they cannot be placed in a rack. Yes, such a server can easily hang on double-sided adhesive tape anchors somewhere under the stairs or in the back room.
Since edge servers are placed outside of secure data centers, they have higher physical security requirements. Protective containers are provided for them:
At the level of working with data in edge servers, disk encryption and secure boot are provided. Encryption itself takes 2-3% of computing power, but edge servers usually use Xeon D processors with a built-in AES acceleration module, which minimizes power losses.
When to Use Edge Servers
With Edge Computing, only data that cannot or is not rational to be processed in a different way is sent to the data center for processing. Thus, edge servers are used when required:
- A flexible approach to security, since in the case of Edge Computing, you can configure the transfer of pre-processed and prepared information to the central data center;
- Protection against loss of information, since if communication with the center is lost, local nodes will accumulate information;
- Savings on traffic, it is achieved due to the processing of the main array of information on the spot.
Edge computing to save traffic
The Danish company Maersk, one of the world's leaders in maritime freight transport, has decided to reduce the fuel consumption of its ships and reduce emissions of pollutants into the atmosphere.
Technology was used to solve this problem. , sensors on the engines and main components of the ship, as well as a local BullSequana Edge server for on-site computing.
Thanks to the sensors, the EcoMain Suite system constantly monitors the condition of the ship's critical components and their deviation from a pre-calculated norm. This allows you to quickly diagnose a malfunction and localize it down to the problem node. Since the telemetry is constantly transmitted "to the center", the service technician can perform the analysis remotely and make recommendations to the onboard crew. And the main question here is how much data and in what volume to transfer to the central data center.
Since it is very problematic to connect cheap wired Internet to a sea container ship, transferring a large amount of raw data to a central server is too expensive. On the BullSequana S200 central server, the general logical model of the ship is calculated, and data processing and direct control are transferred to the local server. As a result, the implementation of this system paid off in three months.
Edge computing to save resources
Another example of edge computing is video analytics. For example, one of the local tasks of the production cycle at Air Liquide, a manufacturer of equipment for technical gases, is quality control of the coloring of gas cylinders. It was carried out manually and was about 7 minutes per cylinder.
To speed up this process, the man was replaced with a block of 7 high-definition video cameras. Cameras film the balloon from multiple angles, generating about 1 GB of video per minute. The video is sent to the edge server BullSequana Edge with Nvidia T4 on board, where a neural network trained to find defects analyzes the stream online. As a result, the average time for inspection was reduced from several minutes to several seconds.
Edge computing in analytics
Attractions at Disneyland are not only fun, but also a complex technical object. So, about 800 different sensors are installed on the Roller Coaster. They constantly send data about the operation of the attraction to the server, and the local server processes this data, calculates the probability of the attraction failing and signals this to the central data center.
Based on these data, the probability of a technical failure is determined and preventive repairs are launched. The attraction continues to work until the end of the working day, and in the meantime, a repair order has already been issued, and workers are promptly repairing the attraction at night.
BullSequana Edge
BullSequana Edge servers are part of a large infrastructure for working with big data, they have already been tested with Microsoft Azure and Siemens MindSphere, VMware WSX platforms and are NVidia NGC / EGX certified. Designed specifically for edge computing, these servers are available in U2 form factor and are available in standard rack, DIN rail, wall and floor mount options.
BullSequana Edge are built on a proprietary motherboard and an Intel Xeon D-2187NT processor. They support the installation of up to 512 GB of RAM, 2 SSDs of 960 GB or 2 HDDs of 8 or 14 TB. They can also install 2 GPU Nvidia T4 16 GB to work with video; Wi-fi, LoRaWAN and 4G modules; up to 2 x 10 Gigabit SFP modules. The servers themselves already have a lid open sensor that is connected to the BMC that controls the IPMI module. It can be configured to automatically turn off power when a sensor is triggered.
Full specifications for BullSequana Edge servers can be found at . If you are interested in the details, we will be happy to answer our questions in the comments.
Source: habr.com